Clinical Utility Of Genotype-Guided Treatment

From a Pharmacogenomic Association to a Clinical Decision

A genotype can be associated with drug response without showing that using the genotype to choose treatment improves patient outcomes. Clinical utility asks a different question: does a strategy that uses genotype information lead to better outcomes than the relevant standard prescribing strategy? The comparison concerns decisions and outcomes, not the strength of an association alone.

The strategies must be specified clearly. A genotype-guided strategy might use a test result to select a drug or dose, while standard prescribing might use the usual clinical information without that result. The relevant evidence compares outcomes under those strategies in the intended patient population. A randomized comparison can estimate the effect of assignment to each strategy; results from an observational comparison require additional assumptions to support a causal interpretation.

For clinical decisions, relative effects are not enough. A relative reduction can correspond to a small or large absolute reduction depending on the event risk under standard care. Harms, testing burdens, costs, and patient preferences also matter. The calculations below use hypothetical data and are not estimates for any real drug or genotype.

Absolute Risks and Treatment Effects

Definition: For a specified outcome and follow-up period, the risk in a strategy group is the proportion of patients who experience that outcome during that period. If $R_G$ is the risk under genotype-guided treatment and $R_S$ is the risk under standard prescribing, the risk difference is $RD=R_G-R_S$. A negative $RD$ indicates a lower risk under genotype-guided treatment; its absolute risk reduction is $ARR=R_S-R_G$, when $R_G<R_S$. The relative risk is $RR=R_G/R_S$, when $R_S>0$.

An absolute risk reduction is usually easier to communicate than a relative reduction because it states how many fewer events occur per patient, or per 100 or 1,000 patients, over a defined period. The number needed to treat (NNT) is the reciprocal of the absolute risk reduction: $NNT=1/ARR$. It is the number of patients who would need to receive the strategy rather than the comparator, over the specified follow-up, to prevent one event on average. NNT is conventionally rounded up to a whole patient.

The same calculations apply to an outcome that is more frequent under the guided strategy. If the risk of a particular harm is $R_G>R_S$, the absolute risk increase is $ARI=R_G-R_S$, and the number needed to harm is $NNH=1/ARI$. These quantities are outcome-specific: preventing treatment failure and causing a serious adverse reaction are not interchangeable events. Their clinical importance must be considered separately or combined using an explicit utility scale.

Worked Example: Absolute Benefit and Harm in a Randomized Strategy Trial

Suppose a hypothetical randomized trial assigns 600 patients to genotype-guided prescribing and 600 to standard prescribing. Follow-up is complete for all patients and lasts 12 months. Treatment failure occurs in 72 guided-strategy patients and 108 standard-strategy patients. A serious drug-related harm occurs in 24 and 12 patients, respectively. Assume randomization was implemented as planned, outcomes were assessed comparably, and the analysis compares patients according to assigned strategy. The resulting estimates describe the effect of assignment in this trial; they do not by themselves guarantee the same effect in every setting.

For treatment failure, the risks are $72/600=0.120$ under genotype guidance and $108/600=0.180$ under standard prescribing. Thus:

ARR=0.180-0.120=0.060,\qquad RR=\frac{0.120}{0.180}=0.667,\qquad NNT=\frac{1}{0.060}=16.7\approx17.

The estimated absolute reduction is 6.0 percentage points over 12 months: about six fewer treatment failures per 100 patients assigned to genotype-guided prescribing. The estimated NNT is 17 patients to prevent one treatment failure on average over 12 months.

For serious drug-related harm, the risks are $24/600=0.040$ and $12/600=0.020$. The absolute increase is $0.040-0.020=0.020$, or 2.0 percentage points, and $NNH=1/0.020=50$. In this trial, the strategy is associated with about two additional serious harms per 100 patients over 12 months, or one additional harm per 50 patients treated. Benefit on one outcome does not erase harm on another.

For illustration, the standard error of the estimated treatment-failure risk difference, using independent binomial groups, is:

SE(RD)=\sqrt{\frac{0.120(1-0.120)}{600} +\frac{0.180(1-0.180)}{600}} =\sqrt{0.000422}=0.0205.

An approximate 95% confidence interval for $RD=R_G-R_S=-0.060$ is $-0.060\pm1.96(0.0205)$, or about $(-0.100,-0.020)$. Under the model used for this interval, the data are compatible with an absolute reduction in treatment failure of roughly 2.0 to 10.0 percentage points. This interval reflects sampling uncertainty, not uncertainty about whether a particular patient will have an event.

Decision Thresholds: When Does a Change in Treatment Make Sense?

A treatment can reduce one outcome and increase another. To decide whether the trade-off is worthwhile, a clinician or patient needs a decision rule that reflects the outcomes’ importance. One simple approach uses loss weights: let $B$ represent the loss assigned to a treatment-failure event and $C$ the loss assigned to a serious-harm event. If the guided strategy reduces failure risk by $ARR$ but increases harm risk by $ARI$, its net benefit on this scale is $B(ARR)-C(ARI)$. A positive value favours the guided strategy on the specified outcomes and weights; a negative value favours standard prescribing. This calculation is useful only if the chosen weights reasonably represent the decision-maker’s preferences and relevant outcomes have not been omitted.

A related decision threshold can be derived for an individual who might receive a treatment alternative because of a genotype result. Let $p$ be the event risk under standard treatment for a patient with a given risk profile, and suppose the alternative has relative risk $r$, where $0\leq r<1$. The absolute event reduction is $p(1-r)$. Let $h$ be the alternative’s excess probability of a specified treatment harm, $B$ the loss from the original event, and $C$ the loss from that harm. On this simplified scale, the alternative is preferred when its expected reduction in event-related loss exceeds its expected added harm-related loss:

B\,p(1-r)>C\,h.

Solving for the standard-treatment risk gives the threshold $p_t=Ch/[B(1-r)]$. Under the stated assumptions, the alternative is favoured when $p>p_t$. This is not a universal threshold: it depends on the treatment effect, harm risk, and value assigned to each outcome. It also assumes those quantities apply to the patient being considered.

Worked Example: Applying a Benefit–Harm Threshold

Suppose, hypothetically, that among patients with an actionable genotype a treatment alternative reduces the risk of a specified treatment failure by 30% relative to standard treatment, so $r=0.70$. Assume it adds a 1.5 percentage-point risk of serious harm, $h=0.015$. For this example, assign the serious harm four times the loss of a treatment failure: $C=4$ and $B=1$. Assume these risks and relative effects are applicable to the patient group under consideration, and that no other outcome or burden affects the choice.

The threshold risk under standard treatment is:

p_t=\frac{C h}{B(1-r)} =\frac{4(0.015)}{1(1-0.70)} =\frac{0.060}{0.30}=0.20.

The implied threshold is a 20% standard-treatment event risk. If a patient’s risk is 25%, the alternative is expected to reduce that event risk by $0.25(1-0.70)=0.075$, or 7.5 percentage points. The weighted benefit is $1(0.075)=0.075$; the weighted added harm is $4(0.015)=0.060$. The net value is $0.075-0.060=0.015$ loss units in favour of the alternative. At a 10% standard-treatment risk, the reduction is $0.10(0.30)=0.030$; the weighted benefit $0.030$ is less than the weighted harm $0.060$, so standard treatment is preferred on these assumptions.

The calculation illustrates why a genotype result alone does not settle the treatment choice. The result may identify a group for whom an alternative has a different effect, but the relevant absolute risks and the decision-maker’s priorities determine whether that difference crosses a clinical threshold.

Comparing Strategies in the Target Population

The value of a testing strategy depends on more than the effect among patients whose genotype changes treatment. It also depends on how common actionable genotypes are, whether prescribing actually changes when indicated, outcomes in other patients, and burdens such as testing cost, turnaround time, and delays. A population comparison should therefore estimate outcomes under the whole genotype-guided strategy and under the relevant standard strategy, not just compare selected genotype groups.

Worked Example: Population Impact and Testing Cost

Consider a hypothetical population of 1,000 patients. An actionable genotype is present in 20%, or 200 patients. Under standard prescribing, treatment failure occurs in 18% of actionable-genotype patients and 10% of the other 800. Under genotype-guided prescribing, the alternative treatment reduces failure risk to 10% in the actionable group; prescribing and outcomes are unchanged for the other group. Assume the trial estimates are transportable to this population, all patients receive the intended strategy, and follow-up and outcome definitions are the same.

Under standard prescribing, the expected number of failures is $200(0.18)+800(0.10)=36+80=116$, a risk of $116/1{,}000=0.116$. Under genotype-guided prescribing, it is $200(0.10)+800(0.10)=20+80=100$, a risk of 0.100. The population absolute reduction is $0.116-0.100=0.016$, or 1.6 percentage points; the corresponding $NNT$ is $1/0.016=62.5$, conventionally about 63 patients. This population NNT is larger than the NNT among actionable-genotype patients because most patients’ prescribing does not change.

Now suppose serious harm risk is 2% under standard treatment in both genotype groups. Under the guided strategy, it remains 2% in the non-actionable group but rises to 3% in the actionable group. Standard care yields $1{,}000(0.02)=20$ serious harms; guided care yields $800(0.02)+200(0.03)=16+6=22$. The excess is two harms per 1,000 patients, or 0.2 percentage points, giving $NNH=1/0.002=500$. These aggregate counts display both the reduction in failure and the increase in serious harm.

Suppose, solely for an illustrative economic comparison, each prevented treatment failure is valued at $2,000, each additional serious harm at $8,000, and testing costs $40 per patient. The estimated monetary benefit from prevented failures is $16($2{,}000)=$32{,}000$; the harm cost is $2($8{,}000)=$16{,}000$; and testing costs $1{,}000($40)=$40{,}000$. The incremental net monetary value is $$32{,}000-$16{,}000-$40{,}000=-$24{,}000$, or $-$24$ per patient. Thus, the guided strategy reduces failures in this example but does not provide positive net monetary value under the specified valuations. Different costs, outcome values, or patient preferences could change that conclusion.

What the Evidence Can—and Cannot—Establish

Clinical utility is an evidence-based comparison of strategies, not a claim that genotypes cause outcomes or that every patient with a particular genotype should receive the same treatment. A randomized strategy trial is especially informative because random assignment can balance measured and unmeasured baseline factors on average. To interpret its results, examine whether assignment was followed, the outcome and follow-up period were clinically relevant, outcome ascertainment was comparable, and the trial population resembles the patients for whom the strategy is proposed. Incomplete follow-up, treatment crossover, or differences in care beyond the genotype-guided decision may complicate interpretation.

Absolute risks and their uncertainty should be reported for important benefits and harms. A small estimated risk difference may be clinically important for a severe outcome, while a larger difference may not matter as much for a mild, reversible one. Confidence intervals help show how imprecisely those differences are estimated. If an interval for a risk difference includes zero, the data do not clearly determine the direction of the effect at that confidence level; this does not establish that the strategies are equivalent. Equivalence requires a prespecified margin and an analysis designed to assess it.

The threshold approach also simplifies real decisions. A genotype test may be inaccurate, unavailable in time, or costly; a treatment choice may affect several outcomes; and patients may differ in how they value those outcomes. If testing can incorrectly classify patients, expected outcomes must account for test performance as well as treatment effects. A useful evaluation makes these assumptions explicit and checks whether the decision changes over plausible ranges of risk, cost, and preference.

Common Mistake: Treating Statistical Evidence as Clinical Value

Wrong reasoning: “The genotype–response association is statistically convincing, so genotype-guided prescribing is clinically useful.” Correction: An association does not establish that using the result improves outcomes compared with standard prescribing. Evaluate the strategy comparison and its absolute benefits, harms, and burdens.
Wrong reasoning: “The relative risk is reduced by 30%, so 30 of every 100 patients benefit.” Correction: A relative reduction is not an absolute percentage-point reduction. If standard-treatment risk is 10%, a 30% relative reduction gives an absolute reduction of $0.10(0.30)=0.03$, or three percentage points.
Wrong reasoning: “There are fewer treatment failures, so the guided strategy is better.” Correction: Compare the harms and other relevant consequences as well. The preferred strategy depends on their absolute frequencies and on how patients and decision-makers value them.
Wrong reasoning: “An NNT of 17 means every seventeenth patient is guaranteed to benefit.” Correction: NNT is the reciprocal of an average risk difference over a defined population and follow-up period. It does not identify which individuals benefit.

Key takeaway: Clinical utility is judged by the outcomes of genotype-guided prescribing compared with standard care. Report absolute risks over a specified period, quantify both benefits and harms, incorporate relevant testing burdens, and make the decision threshold and its value assumptions explicit.

Check Your Understanding

Use the definitions and decision logic above to answer these questions.

A strategy reduces an outcome risk from 15% to 12% over one year. Calculate the absolute risk reduction, relative risk, and NNT, and interpret each in context.
Why can an association between genotype and drug response fail to establish clinical utility?
In a randomized strategy trial, what does an intention-to-treat comparison estimate, and what assumptions or limitations should be considered when applying it to a different patient population?
If a guided strategy reduces treatment failure but increases serious drug harm, what additional information is needed before deciding which strategy is preferable?
In the threshold model, explain how an increase in the excess harm risk $h$ affects the threshold $p_t$, holding the other quantities fixed.